On Sun, 2010-03-14 at 18:33 +0200, Maxim Levitsky wrote: > Hi, > > I have some interesting results in regard to long transmission blackouts > I experience with my ath5k based wifi card (AR2424 inside aspire one). > > This is very old issue, and unfortunately I didn't have time to debug it > properly. Fortunately, and strangely card works OK, if only used to surf > the webs and download files. > > This is standard testcase (don't know why nc sometimes exits, probably > due to interrupted syscall): > > This is run on the ath5k hosting machine. At same time network is quiet, > and channel I use (#1) is free of neighbours APs. > > Open AP is used, no networkmanager is running, so nothing could trigger > the scan of wireless networks, and indeed the printks confirmed that > card isn't reset randomly. > > I do: > > while true ; do nc -u $SERVER 1 < /dev/zero ; done > > > After a while (no resets meanwhile), transmission stops. > > After adding some strategic printks, I see that card sends > AR5K_INT_RXORN interrupt, which is very strange, because all the card > does is transmitting. > > On purpose I disabled the card reset that follows the AR5K_INT_RXORN, > and see that transmission halts before the reset. > > The reset usually restores functionality, but often it is followed by > another bath of AR5K_INT_RXORN errors. > > Sometimes however, after the reset, nothing happens, nothing > transmitted, until another reset is done. This also usually triggers > disassociation, and long stalls on all TCP connections. enough said... > > > To be sure I merged the git tip of the wireless-testing tree only to > find out that nothing changed. I use vanilla git tree from kernel.org, > but this issue is so old that the above isn't relevant much. > > When I say 'high load' I probably overestimate it. I could trigger that > bug by uploading to affected machine. The stream of tcp ACKs is enough > to trigger this (about 80 Kbytes/s if I remember correctly) > > Maybe deterministic timing between packets is what triggers this bug. > What I know that a normal download from the web at ~300 Kbytes/s doesn't > trigger that bug. > > This is output of my hand crafted printks: > > <4>[ 2489.249349] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.249882] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.250435] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.250881] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.251318] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.251768] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.252117] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.252640] ath5k_hw_get_isr: ISR: 0x00000080 IMR: 0x800824b5 > <4>[ 2489.254196] ath5k_hw_get_isr: ISR: 0x00000001 IMR: 0x800824b5 > <4>[ 2490.586420] ath5k_hw_get_isr: ISR: 0x00000020 IMR: 0x800824b5 > <4>[ 2490.586432] ath5k: reset due to AR5K_INT_RXORN > <4>[ 2490.586453] ath5k_hw_get_isr: ISR: 0x00000020 IMR: 0x800824b5 > <4>[ 2490.586461] ath5k: reset due to AR5K_INT_RXORN > <4>[ 2490.586479] ath5k_hw_get_isr: ISR: 0x00000020 IMR: 0x800824b5 > <4>[ 2490.586487] ath5k: reset due to AR5K_INT_RXORN > <4>[ 2490.586505] ath5k_hw_get_isr: ISR: 0x00000020 IMR: 0x800824b5 > > Best regards, > Maxim Levitsky >
Luis, I think this is some internal chip problem. Maybe you can look at your internal documentation and see what is going on. Meanwhile, I verified that phy caliration interval doesn't affect this problem. It pretty much is as simple, as this: While transmitting, card randomly stops transmitting (in fact it stops issuing any IRQs) and then it starts to bomb out the AR5K_INT_RXORN (In above trace you see it gave RXOK, before RXORN, but this doesn't usually happen.) A reset doesn't help always. I really don't know what to do next. Also note that for completeness sake I tried the ANI patch, the patch in this thread. I also tried once MadWifi, and had same results. Best regards, Maxim Levitsky _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
